usage vs acquisition in language change andrew wedel and clay beckner language as a complex system...
Post on 22-Dec-2015
213 views
TRANSCRIPT
![Page 1: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/1.jpg)
Usage vs Acquisition in Language Change
Andrew Wedel and Clay Beckner
Language as a Complex System Workshop
University of Arizona, 2008
![Page 2: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/2.jpg)
Change in a steady-state population requires turnover
• Can be thought of as information loss + information gain.
• Bottlenecks in biological evolution
![Page 3: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/3.jpg)
Kirby: Language acquisition is an information bottleneck
![Page 4: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/4.jpg)
Acquisition-based model of compositionality vs. idiosyncrasy
• 2 dimensional meaning space with 25 bimorphemic words– a0-4 x b0-4 = 25 meanings– Each meaning is associated with a string of characters.
• Speaker utters 50 randomly chosen meanings to learner, who stores mappings between strings and meanings.
• Learner becomes new speaker, and utters another 50 random meanings to new learner.– If the meaning was encountered in the learning phase, the
corresponding string is uttered.– If not, composition is attempted from other words sharing meaning
components.– Small chance of deleting a character
![Page 5: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/5.jpg)
Frequency differences of meaning expressions
![Page 6: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/6.jpg)
Sample Results
• Intended take home message: high frequency enables idiosyncrasy to survive the acquisition bottleneck.
![Page 7: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/7.jpg)
Change through drift
• Information is lost and gained in any population with turnover, whether abrupt or gradual.– Drift in biological populations– Drift in populations within categories through usage
Language change through information loss/gain could arise in principle in acquisition OR in usage.
![Page 8: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/8.jpg)
Illustrate functional equivalence with Kirby-esque model
• Translate Kirby model into an exemplar format:– Words comprise 2 morphemes of 2 characters
each.• {a1, a2, a3, a4} + {b1, b2, b3, b4}
– Listener is privy to the speaker’s meaning, and stores all percepts as exemplars of that meaning category
• Generalizations about string-morpheme mappings are passively stored within the total exemplar set, rather than as explicit rules.
![Page 9: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/9.jpg)
Illustrate functional equivalence with Kirby-esque model
• Translate Kirby model into an exemplar format:– Provide two pathways for production of a meaning:
• Holistic: An exemplar is chosen at random from the set for the current meaning and produced.
• Compositional: The most frequent character string types for each morpheme in the current meaning are concatenated and produced.
– Bias choice of pathway by frequency• Highly frequent forms rarely use the compositional pathway.
• Very infrequent forms use the compositional pathway relatively more.
![Page 10: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/10.jpg)
Bottleneck versus Drift Models
• Bottleneck:– Two agents, conceived as teacher and learner.– Teacher produces 200 outputs for a naive
learner, who stores each as an exemplar in the corresponding meaning category.
– Learner becomes teacher to new naive learner.• Teacher’s choice of holistic versus compositional
routes for each production influenced by the number of times the teacher heard each form as a learner.
![Page 11: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/11.jpg)
Bottleneck versus Drift Models
• Drift:– Two agents conceived as continuously
conversing to each other.– Speaker produces an output which is stored
as a new exemplar by the listener, displacing an old exemplar.
• Speaker’s choice of holistic versus compositional routes for each production influenced by the frequency with which that meaning has been previously heard.
![Page 12: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/12.jpg)
Results
• Run 200 independently seeded simulation runs in both drift and bottleneck mode.– Each run involves 40,000 productions
• Calculate a regularity score for the most common exemplar in each meaning category– 1 if morphemes match the highest type frequency
in the lexicon.– 0 otherwise.
• Average scores for each meaning over 200 runs.
![Page 13: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/13.jpg)
ResultsRegularity vs Frequency in Bottleneck and Drift Models
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
Frequency (1 = low, 16 = high)
Regularity
Drift
Bottleneck
![Page 14: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/14.jpg)
Interim conclusion
• Change through information turnover in acquisition versus usage is not qualitatively different at this level of abstraction.
• What is qualitatively different between acquisition and usage? – Rate of category label creation/loss
![Page 15: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/15.jpg)
Category creation versus category usage
• Category creation involves abstraction of category labels from distributions in input data.– Language acquisition
• Usage involves mapping data to and from existing category labels.– Language use
![Page 16: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/16.jpg)
Categories contain variants/letter ‘a’/
[a A A a a å a a]
Two qualitatively different kinds of change given this conception:Change in category labels, or change in category contents.
![Page 17: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/17.jpg)
Influence of existing categories on present behavior
• Wide array of evidence that perception and production are influenced by preexisting category structure.– Perception:
• Categorical perception (?)• Perceptual magnet effect (Kuhl 1991)
– Production:• Motor patterns (e.g., Zanone & Kelso 1997)• Language change patterns (cf. ‘Choice’,
Blevins 2004)
![Page 18: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/18.jpg)
Category creation versus category usage
• Acquisition and usage contribute qualitatively distinct pathways of language change.– Acquisition
• Changes in category system
– Usage• Changes in category contents
• Each constrains the other.
![Page 19: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/19.jpg)
Illustrate with contrast-maintenance simulation
• Horizontal, usage phase– Agents produce forms with reference to existing
sound categories• Sound outputs are an average, with sampling biased
toward within-category exemplars.
• Vertical, acquisition phase– Agents abstract new sound categories on the
basis of the distribution of perceived sounds.
![Page 20: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/20.jpg)
Expected properties of this architecture
• Change in category contents in the horizontal phase should be influenced by the existing sound category labels
• Change in the set of sound category labels should be influenced by changes in category contents.
![Page 21: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/21.jpg)
Horizontal phase: two sound categories
a b
![Page 22: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/22.jpg)
Horizontal phase: one sound category
a
![Page 23: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/23.jpg)
Vertical phase
• Depending on the breadth of a bimodal distribution from the preceding horizontal phase, it may be acquired as one or two categories in the vertical phase.– with distinct repercussions on the trajectory
of change in the following horizontal phase.
![Page 24: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/24.jpg)
The interdependency of category label abstraction and category use
• Two experiments:1. Keep the number of total vertical
acquisition steps constant, but vary the number of horizontal usage steps between.
2. Keep the number of total horizontal usage steps constant, but vary the number of vertical acquisition steps.
![Page 25: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/25.jpg)
Experiment structure
• Six word categories• Seed with a sound distribution that will
be initially acquired into nine categories.– With six word categories each comprising
two sounds, three sound categories is the most stable state in this simulation architecture.
• Look at rate of relaxation from nine towards three sound categories.
![Page 26: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/26.jpg)
1: Acquisition steps constant, usage steps vary
Usage versus Acquisition in change:varying the number of usage steps
0
1
2
3
4
5
6
7
8
9
10
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Generations (Acquisition steps)
Number of sound categories
10
20
40
80
Each point is the average of 10 independent runs
![Page 27: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/27.jpg)
2: Acquisition steps vary, usage steps constant
Usage versus acquisition in change: Varying the number of acquisition steps
0
1
2
3
4
5
6
7
8
9
10
0 80 1602403204004805606407208008809601040
Usage steps
Number of sound categories
20
40
80
160
320
Each point is the average of 10 independent runs
![Page 28: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/28.jpg)
Conclusions
• Usage versus acquisition may not be a particularly useful apposition in understanding language change given that they do not refer to well-distinguished mechanisms.– Acquisition involves usage– At least some kinds of language acquisition
continue throughout life
![Page 29: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/29.jpg)
Conclusions
• A more useful distinction may lie between change in category labels versus change in category mappings– Roughly correspond to frequent types of
changes that occur in ‘acquisition’ versus ‘usage’.
![Page 30: Usage vs Acquisition in Language Change Andrew Wedel and Clay Beckner Language as a Complex System Workshop University of Arizona, 2008](https://reader034.vdocuments.us/reader034/viewer/2022051619/56649d7d5503460f94a5fe92/html5/thumbnails/30.jpg)
Conclusions• We illustrated interaction of both types of
change in a simulation instantiating two experimentally-supported properties of language:
1. Pathways of change in usage are influenced by the existing set of categories.
2. Abstraction of category labels is influenced by the input distribution -- which evolves through usage in the previous generations.